Research On Short-term Power Load Forecasting Of Smart Grid Based On Deep Learning

Deep learning is a research hotspot in the field of artificial intelligence.With its powerful automatic feature extraction capability,it can quickly analyze large amounts of data and has powerful data fitting ability,and its application in smart grid has been deepening in recent years.The development of artificial intelligence can be summarized into two aspects:(1)With the advancement of smart grid construction,the amount of data generated by its planning,operation and maintenance has increased in the form of indicators;(2)The improvement of deep learning theory has also laid a good foundation for its application in the smart grid.Short-term power load forecasting not only plays a vital role in power cost budgeting,power dispatching,but also plays a vital role in managing the safety of power transmission systems.Based on the latest development research results of deep learning,and according to the actual load data,this paper conducts related research on short-term power load forecasting.The main contents include:(1)The research status of short-term load forecasting is analyzed and the deep learning is applied to the load forecasting of smart grid.The time-varying nonlinearity and stochastic uncertainty of smart grids make it difficult for traditional modeling and analysis methods to comprehensively and thoroughly analyze the characteristics of smart grid load data in the new form.The introduction of deep learning technology in artificial intelligence can not only reduce the difficulty of extracting load data features from the smart grid,but also comprehensively mine highdimensional complex data to make up for a series of traditional problems such as poor generalization ability,insufficient training and so on.(2)A power load forecasting model based on deep belief network is presented.The traditional prediction method does not have the ability to convert and process large data samples,so that the learning ability is limited.When the prediction accuracy reaches a certain height,it is difficult to improve.With the increase of power consumption,the traditional load forecasting method can not meet the actual requirements for higher and higher prediction accuracy.A deep belief network prediction model is proposed.This model is an efficient and unsupervised deep network model.By combining low-level features,a more abstract high-level representation is formed to discover the distributed feature representation of data.By learning a nonlinear network structure,the prediction model realizes complex function approximation,characterizes the distributed representation of input data,and concentrates the ability of the essential characteristics of the data set from a small number of samples.(3)A feature extraction prediction model based on evolutionary deep learning is proposed.At present,there is no strict theoretical guidance on the number of hidden layers and the number of hidden layer units.If it is not properly set,it may bring problems such as insufficient precision or long processing time.Deep learning is very sensitive to the setting of parameters related to hidden layers.The setting of network model parameters is basically based on experience.In response to these problems,the core ideas of genetic algorithms and evolutionary strategies are integrated into the deep belief network to simplify the network structure and construct a deep learning model with good performance.